Clustering::MPP Namespace Reference

functions related to "Most Probable Path"-clustering More...

Typedefs
using	SparseMatrixF = boost::numeric::ublas::mapped_matrix< float >
	BOOST implementation of a sparse matrix for floats.
using	Neighborhood = Clustering::Tools::Neighborhood
	Neighborhood per frame.

Functions
SparseMatrixF	read_transition_probabilities (std::string fname)
	read (row-normalized) transition matrix from file
SparseMatrixF	transition_counts (std::vector< std::size_t > trajectory, std::vector< std::size_t > concat_limits, std::size_t n_lag_steps, std::size_t i_max)
SparseMatrixF	weighted_transition_counts (std::vector< std::size_t > trajectory, std::vector< std::size_t > concat_limits, std::size_t n_lag_steps)
SparseMatrixF	row_normalized_transition_probabilities (SparseMatrixF count_matrix, std::set< std::size_t > microstate_names)
	compute transition matrix from counts by normalization of rows
SparseMatrixF	updated_transition_probabilities (SparseMatrixF transition_matrix, std::map< std::size_t, std::size_t > sinks, std::map< std::size_t, std::size_t > pops)
	update transition matrix after lumping states into sinks
std::map< std::size_t, std::size_t >	single_step_future_state (SparseMatrixF transition_matrix, std::set< std::size_t > cluster_names, float q_min, std::map< std::size_t, float > min_free_energy)
std::map< std::size_t, std::vector< std::size_t > >	most_probable_path (std::map< std::size_t, std::size_t > future_state, std::set< std::size_t > cluster_names)
std::map< std::size_t, std::size_t >	microstate_populations (std::vector< std::size_t > clusters, std::set< std::size_t > cluster_names)
	compute cluster populations
std::map< std::size_t, float >	microstate_min_free_energy (const std::vector< std::size_t > &clustering, const std::vector< float > &free_energy)
std::map< std::size_t, std::size_t >	path_sinks (std::vector< std::size_t > clusters, std::map< std::size_t, std::vector< std::size_t >> mpp, SparseMatrixF transition_matrix, std::set< std::size_t > cluster_names, float q_min, std::vector< float > free_energy)
std::vector< std::size_t >	lumped_trajectory (std::vector< std::size_t > trajectory, std::map< std::size_t, std::size_t > sinks)
std::tuple< std::vector< std::size_t >, std::map< std::size_t, std::size_t >, SparseMatrixF >	fixed_metastability_clustering (std::vector< std::size_t > initial_trajectory, SparseMatrixF trans_prob, float q_min, std::vector< float > free_energy)
	run clustering for given Q_min value
void	main (boost::program_options::variables_map args)
	MPP clustering control function and user interface. More...

Detailed Description

functions related to "Most Probable Path"-clustering

This module contains all function for dynamical clustering. In contrast to density-based clustering, can it only be applied to previously clustered trajectories. The idea is to create, based on a microstate input, a coarse-grained model (macrostates). The most probable path depends strongly on the selected timescale (mpp time). If the input was dynamically cored, the mpp time needs to be greater than the coring time.

Function Documentation

lumped_trajectory()

std::vector< std::size_t > Clustering::MPP::lumped_trajectory	(	std::vector< std::size_t >	trajectory,
		std::map< std::size_t, std::size_t >	sinks
	)

lump states based on path sinks and return new trajectory. new microstates will have IDs of sinks.

Definition at line 400 of file mpp.cpp.

main()

void Clustering::MPP::main

(

boost::program_options::variables_map

args

)

MPP clustering control function and user interface.

Parameters

input	input file with microstate trajectory
basename	name format for output files
lagtime	lag for transition estimation in units of frame numbers
qmin-from	lower limit for metastability (Q_min)
qmin-to	upper limit for metastability (Q_min)
qmin-step	stepping for metastability (Q_min)
concat-nframes	no. of frames per trajectory.
concat-limits	length of concated trajectories.

Returns

void

Note

Lagtime should be greater than the coring time/ smallest timescale.

Definition at line 487 of file mpp.cpp.

microstate_min_free_energy()

std::map< std::size_t, float > Clustering::MPP::microstate_min_free_energy	(	const std::vector< std::size_t > &	clustering,
		const std::vector< float > &	free_energy
	)

assign every state the lowest free energy value of all of its frames.

Definition at line 320 of file mpp.cpp.

most_probable_path()

std::map< std::size_t, std::vector< std::size_t > > Clustering::MPP::most_probable_path	(	std::map< std::size_t, std::size_t >	future_state,
		std::set< std::size_t >	cluster_names
	)

for every state, compute most probable path by following the 'future_state'-mapping recursively

Definition at line 288 of file mpp.cpp.

path_sinks()

std::map< std::size_t, std::size_t > Clustering::MPP::path_sinks	(	std::vector< std::size_t >	clusters,
		std::map< std::size_t, std::vector< std::size_t >>	mpp,
		SparseMatrixF	transition_matrix,
		std::set< std::size_t >	cluster_names,
		float	q_min,
		std::vector< float >	free_energy
	)

compute path sinks, i.e. states of highest metastability, and lowest free energy per path. these sinks will be states all other states of the given path will be lumped into.

Definition at line 337 of file mpp.cpp.

single_step_future_state()

std::map< std::size_t, std::size_t > Clustering::MPP::single_step_future_state	(	SparseMatrixF	transition_matrix,
		std::set< std::size_t >	cluster_names,
		float	q_min,
		std::map< std::size_t, float >	min_free_energy
	)

compute immediate future (i.e. without lag) of every state from highest probable transitions; exclude self-transitions.

Definition at line 234 of file mpp.cpp.

transition_counts()

SparseMatrixF Clustering::MPP::transition_counts	(	std::vector< std::size_t >	trajectory,
		std::vector< std::size_t >	concat_limits,
		std::size_t	n_lag_steps,
		std::size_t	i_max = 0
	)

count transitions from one to the other cluster with certain lag and return as count matrix (row/col := from/to)

Definition at line 78 of file mpp.cpp.

weighted_transition_counts()

SparseMatrixF Clustering::MPP::weighted_transition_counts	(	std::vector< std::size_t >	trajectory,
		std::vector< std::size_t >	concat_limits,
		std::size_t	n_lag_steps
	)

same as 'transition_counts', but with reweighting account for differently sized trajectory chunks (as given by concat_limits)

Definition at line 113 of file mpp.cpp.